Learning Model Generation for Catheter Image Edge Classification

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Images acquired using an image-acquiring catheter often have defects, making it difficult to correctly visualize the structure of luminal organs due to missing information, particularly when the lumen region extends beyond the image edge.

Innovation Solution

A learning model generation method that classifies pixels in two-dimensional images into living tissue, lumen, and extra-luminal regions, determines if the lumen region reaches the image edge, and creates a division line to separate the lumen region into relevant areas, generating classification data to aid in understanding and recording the images for training databases.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of information

If image-acquiring catheter is inserted into luminal organ to acquire images, then in-vivo structure can be visualized, but image defects occur when lumen region reaches image edge causing loss of information

Engineering Contradiction:
Improvemissing information in imageVSAvoidcorrect visualization of luminal organ structure
Core Design Contradiction:
Loss of informationVSReliability

Solution Approach 1:

The patent applies preliminary action by creating a division line in the training data before machine learning processing. When the lumen region reaches the image edge, the division line pre-separates the lumen region into first and second regions, assigning different labels (lumen vs. extra-luminal) to prevent information loss. This preliminary classification enables the learning model to correctly interpret edge regions without requiring manual correction during actual imaging.

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If lumen region is classified uniformly without division, then classification process is simple, but accuracy decreases when lumen reaches image edge

Engineering Contradiction:
Improveclassification accuracyVSAvoidclassification process complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies local quality by making the classification approach adaptive to local image conditions. The system determines whether the lumen region reaches the image edge and applies different classification strategies accordingly: uniform classification for non-edge cases and division-line-based classification for edge cases. This localized adaptation maintains simplicity where possible while improving accuracy where needed.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The division line is created preliminarily in the training data generation phase, not during real-time image processing. This preliminary action embeds the complexity handling into the training process, allowing the learning model to automatically apply the appropriate classification logic during actual use without adding real-time computational burden.

Inventive Principle:
Principle #10Preliminary action

3Measurement precision

If manual correction is performed for edge cases, then classification accuracy improves, but time consumption and operation complexity increase

Engineering Contradiction:
Improveclassification accuracyVSAvoidmanual correction time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent implements self-service by enabling the learning model to automatically handle edge cases without requiring manual correction. During training, the model learns from data that includes pre-processed division lines and appropriate labels. During actual imaging, the trained model autonomously determines whether to apply division-based classification, eliminating the need for user intervention and significantly reducing time consumption.

Inventive Principle:
Principle #25Self-service

4Loss of information

If division line is created for every lumen region, then information accuracy is maintained, but processing complexity and computational load increase

Engineering Contradiction:
Improveinformation completenessVSAvoidprocessing complexity
Core Design Contradiction:
Loss of informationVSDevice complexity

Solution Approach 1:

The patent applies partial action by creating division lines only when necessary - specifically when the lumen region reaches the image edge. The system evaluates each image case and applies the division-line-based classification only to images where it is needed, rather than uniformly to all images. This selective approach maintains information accuracy while minimizing unnecessary processing complexity.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentUS20240221366A1Learning model generation method, image processing apparatus, information processing apparatus, training data generation method, and image processing method
Publication Date: 2024.07.04 TERUMO KK
  • US20240221366A1 patent drawing
  • US20240221366A1 patent drawing
  • US20240221366A1 patent drawing

AI summary

A learning model generation method for generating a learning model that aids understanding of an image acquired with an image-acquiring catheter. The learning model generation method includes: creating a division line that divides a lumen region into a first region into which the image-acquiring catheter is inserted and a second region reaching an edge of a two-dimensional image, when it is determined that the lumen region reaches an edge of the two-dimensional image; creating second classification data in which a probability of being the lumen region and a probability of being an extra-luminal region are allocated; recording the two-dimensional image associated with the second classification data in a training database; and generating a learning model that outputs third classification data in which an input two-dimensional image is classified into a plurality of regions including a living tissue region, the lumen region, and an extra-luminal region, by machine learning.